Jiacheng Liu

Jiacheng Liu, Chang Zou, Yuanhuiyi Lyu et al.

Diffusion Transformers (DiT) have revolutionized high-fidelity image and video synthesis, yet their computational demands remain prohibitive for real-time applications. To solve this problem, feature caching has been proposed to accelerate diffusion models by caching the features in the previous timesteps and then reusing them in the following timesteps. However, at timesteps with significant intervals, the feature similarity in diffusion models decreases substantially, leading to a pronounced increase in errors introduced by feature caching, significantly harming the generation quality. To solve this problem, we propose TaylorSeer, which firstly shows that features of diffusion models at future timesteps can be predicted based on their values at previous timesteps. Based on the fact that features change slowly and continuously across timesteps, TaylorSeer employs a differential method to approximate the higher-order derivatives of features and predict features in future timesteps with Taylor series expansion. Extensive experiments demonstrate its significant effectiveness in both image and video synthesis, especially in high acceleration ratios. For instance, it achieves an almost lossless acceleration of 4.99$\times$ on FLUX and 5.00$\times$ on HunyuanVideo without additional training. On DiT, it achieves $3.41$ lower FID compared with previous SOTA at $4.53$$\times$ acceleration. %Our code is provided in the supplementary materials and will be made publicly available on GitHub. Our codes have been released in Github:https://github.com/Shenyi-Z/TaylorSeer

Jiacheng Liu, Peiliang Cai, Qinming Zhou et al. · cmu

The application of diffusion transformers is suffering from their significant inference costs. Recently, feature caching has been proposed to solve this problem by reusing features from previous timesteps, thereby skipping computation in future timesteps. However, previous feature caching assumes that features in adjacent timesteps are similar or continuous, which does not always hold in all settings. To investigate this, this paper begins with an analysis from the frequency domain, which reveal that different frequency bands in the features of diffusion models exhibit different dynamics across timesteps. Concretely, low-frequency components, which decide the structure of images, exhibit higher similarity but poor continuity. In contrast, the high-frequency bands, which decode the details of images, show significant continuity but poor similarity. These interesting observations motivate us to propose Frequency-aware Caching (FreqCa) which directly reuses features of low-frequency components based on their similarity, while using a second-order Hermite interpolator to predict the volatile high-frequency ones based on its continuity. Besides, we further propose to cache Cumulative Residual Feature (CRF) instead of the features in all the layers, which reduces the memory footprint of feature caching by 99%. Extensive experiments on FLUX.1-dev, FLUX.1-Kontext-dev, Qwen-Image, and Qwen-Image-Edit demonstrate its effectiveness in both generation and editing. Codes are available in the supplementary materials and will be released on GitHub.

Shikang Zheng, Guantao Chen, Qinming Zhou et al.

Diffusion Transformers offer state-of-the-art fidelity in image and video synthesis, but their iterative sampling process remains a major bottleneck due to the high cost of transformer forward passes at each timestep. To mitigate this, feature caching has emerged as a training-free acceleration technique that reuses or forecasts hidden representations. However, existing methods often apply a uniform caching strategy across all feature dimensions, ignoring their heterogeneous dynamic behaviors. Therefore, we adopt a new perspective by modeling hidden feature evolution as a mixture of ODEs across dimensions, and introduce HyCa, a Hybrid ODE solver inspired caching framework that applies dimension-wise caching strategies. HyCa achieves near-lossless acceleration across diverse domains and models, including 5.55 times speedup on FLUX, 5.56 times speedup on HunyuanVideo, 6.24 times speedup on Qwen-Image and Qwen-Image-Edit without retraining.

Jiacheng Liu, Xinyu Wang, Yuqi Lin et al.

Diffusion Models have become a cornerstone of modern generative AI for their exceptional generation quality and controllability. However, their inherent \textit{multi-step iterations} and \textit{complex backbone networks} lead to prohibitive computational overhead and generation latency, forming a major bottleneck for real-time applications. Although existing acceleration techniques have made progress, they still face challenges such as limited applicability, high training costs, or quality degradation. Against this backdrop, \textbf{Diffusion Caching} offers a promising training-free, architecture-agnostic, and efficient inference paradigm. Its core mechanism identifies and reuses intrinsic computational redundancies in the diffusion process. By enabling feature-level cross-step reuse and inter-layer scheduling, it reduces computation without modifying model parameters. This paper systematically reviews the theoretical foundations and evolution of Diffusion Caching and proposes a unified framework for its classification and analysis. Through comparative analysis of representative methods, we show that Diffusion Caching evolves from \textit{static reuse} to \textit{dynamic prediction}. This trend enhances caching flexibility across diverse tasks and enables integration with other acceleration techniques such as sampling optimization and model distillation, paving the way for a unified, efficient inference framework for future multimodal and interactive applications. We argue that this paradigm will become a key enabler of real-time and efficient generative AI, injecting new vitality into both theory and practice of \textit{Efficient Generative Intelligence}.

Xinfeng Xia, Jiacheng Liu, Xiaofeng Hou et al.

Mixture-of-Experts (MoE) models, the state-of-the-art in large-scale AI, achieve high quality by sparsely activating parameters. However, their reliance on routing between a few monolithic experts via a top-k mechanism creates a "quality cliff", offering only a few coarse-grained operating points. This inflexibility forces a difficult trade-off between cost and quality, preventing adaptation to diverse Service Level Objectives (SLOs) and leading to significant resource over-provisioning. This paper introduces MoE-Prism, a model-system co-design that transforms rigid MoE models into elastic services. Our methodology is divided into two phases. First, an \emph{Offline Refactoring Engine} systematically deconstructs monolithic experts into fine-grained "sub-experts." This engine employs a partitioning optimization solver that uses a metaheuristic-based approach to group neurons, preserving functional locality without requiring retraining. Second, an \emph{Online Scheduling Engine} leverages this new elasticity through QoS-aware scheduling. It implements specialized policies to solve complex system problems, including maximizing throughput in cloud deployments and managing latency-optimized offloading for memory-constrained devices. Our evaluation across three different MoE models shows that MoE-Prismprovides over 4 times more distinct, stable operating points than the baseline. This allows an AI service to dynamically improve throughput by up to 19.9\% under a strict latency budget or reduce latency by up to 10.36\% under limited resources. MoE-Prism provides the critical "control knob" to bridge the model-system gap, enabling the next generation of adaptive, efficient, and QoS-aware AI services.